Presentia documentation

Presentia is used mostly from the dashboard: create a deck from a prompt, a document, a dataset or markdown; fine-tune it in the editor (with an AI assistant); export to PPTX or PDF. This page covers the fork’s own features and the API; the day-to-day editor works as you would expect from its upstream.

Markdown to deck (Gamma mode)

The Markdown page has a full editor — toolbar, live preview that shows exactly how your text splits into cards, card counter, drag & drop — plus template, language, image source/style and export pickers. Every section split by --- or by #/## headings becomes a card.

`text_mode`	What happens
`preserve`	Your text goes into the deck verbatim — the AI only picks layouts and generates images.
`condense`	The AI summarizes each card while keeping its structure.
`generate`	The AI rewrites and expands each card.

The same power is available over the API:

curl -X POST http://localhost:5001/api/v1/ppt/presentation/generate-from-markdown \
  -H "Content-Type: application/json" \
  -d '{
        "markdown": "# Q3 Review\nIntro...\n\n---\n\n## Results\n- Revenue up 20%",
        "text_mode": "preserve",
        "template": "general",
        "export_as": "pptx"
      }'

Charts from real data

POST /api/v1/ppt/presentation/generate-from-data builds a deck from a CSV, TSV or JSON dataset (multipart: file, plus content, n_slides, language, template, instructions, export_as). The anti-hallucination guard checks every chart figure against the dataset — exact values or exact column aggregates (sum, average, min, max, count) only. Wrong numbers get one retry with feedback and are rejected if the model insists.

curl -X POST http://localhost:5001/api/v1/ppt/presentation/generate-from-data \
  -F "file=@monthly_summary.csv" \
  -F "content=Monthly reconciliation report" \
  -F "n_slides=6" -F "language=English" -F "export_as=pdf"

Datasets are expected pre-aggregated; the row cap is DATASET_MAX_ROWS (default 200). The default template is Report, which carries the chart layouts.

Costs and models

Two dashboard pages the upstream doesn’t have:

Models — the catalog of text and image models with curated quality dots, blended price per million tokens, availability computed from your keys and recommendation badges. Click a card to switch; with an OpenRouter key nearly everything lights up at once.
Costs — summary cards (calls, tokens in/out, estimated cost) and per-deck breakdowns by stage, slide and model, plus a provider comparison table. Pricing comes from a versioned catalog in the repo, so cost history is auditable.

Escriba Suite integration

Presentia works standalone — without any of these variables it behaves exactly like vanilla Presenton. Each integration is opt-in and independent; the repo’s docker-compose.yml ships commented service blocks for all of them.

Escriba — document parsing with OCR. ESCRIBA_ENABLED=true + ESCRIBA_URL delegate document → markdown conversion to an Escriba service (scanned PDFs, rotated pages, images, 20+ formats). If Escriba is down or fails on a file, Presentia falls back automatically to its local parsers.
Anonimal — PII anonymization before the LLM. ANONIMAL_ENABLED=true + ANONIMAL_URL anonymize user prompts and extracted document text before they travel to the LLM provider. Fail-closed: if the service doesn’t answer, generation stops with a clear error instead of sending raw PII. Original text never leaves your host.
Searchgirl — private web search. WEB_SEARCH_PROVIDER=searchgirl + SEARCHGIRL_BASE_URL ground generations through your own metasearch instance instead of a third-party API (with an optional Bearer token).

Templates, themes and languages

Built-in template families (General, Modern, Standard, Swift, Report with charts, Institucional in es-AR), custom templates generated from your own PPTX, a full theme editor (colors, fonts, logo) and per-template style_instructions that shape the writing tone. The whole interface ships in 7 languages — English, Español, Français, Português, Italiano, 中文, 日本語 — switchable from the sidebar and remembered per browser.

Development

docker compose up -d --build development   # hot reload for both servers
cd servers/fastapi && python -m pytest tests/unit   # the fork adds 100+ unit tests

View on GitHub